Federated RL SEM Defect Detection With Private Model Aggregation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current defect detection tools in semiconductor inspection, particularly those using scanning electron microscopy (SEM), face limitations due to rule-based techniques, leading to misclassification of defects and increased engineering time.
Innovation Solution
A federated, reinforcement-learning (RL) machine learning system is introduced for automatic defect detection and classification. This system involves a central model server providing an initial trained model to clients, who update it based on local data without sharing training data, and then aggregate these updates to form a global model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If rule-based techniques are used for defect classification and detection, then the system is simple to implement, but misclassification of defects occurs and engineering time increases
Solution Approach 1:
The patent replaces rule-based techniques (mechanical/systematic approach) with machine learning models that learn patterns from data. The ML models automatically classify defects based on learned features rather than predefined rules, improving accuracy while reducing manual engineering time for classification.
Solution Approach 2:
The patent changes the approach from fixed rule-based parameters to dynamic parameters learned from training data. The machine learning models adapt their classification parameters based on the specific characteristics of defects in the training set, allowing more accurate and efficient classification without manual rule tuning.
2Reliability
If local training data is shared across clients to improve model accuracy, then defect detection accuracy improves, but data privacy concerns arise
Solution Approach 1:
The patent segments the training process into local and global components. Each client trains models locally on their own data, keeping data private. Only model parameters (not raw data) are shared with the central server for aggregation. This segmentation allows accuracy improvement through data diversity while maintaining data privacy.
Solution Approach 2:
The patent introduces a central server as an intermediary that aggregates model parameters from multiple clients without accessing their raw training data. The server combines learned parameters from different clients to create an improved global model, enabling collaborative learning while preserving data privacy through this intermediary layer.
3Measurement precision
If conventional optical tools are used for inspection, then the inspection process is faster, but measurement precision is insufficient for sub-40 nm structures
Solution Approach 1:
The patent applies preliminary action by using machine learning models to pre-process and pre-classify defects in SEM images. The ML models quickly identify potential defects and their characteristics before detailed analysis, enabling faster processing of high-resolution SEM data while maintaining measurement precision for sub-40 nm structures.
Data Source
AI summary
A federated machine learning method is provided. The method includes providing, from a central model server, an initial trained machine learning (ML) model to a plurality of clients as a respective local ML model. The initial trained ML model is configured to identify defect features from scanning electron microscopy (SEM) images. The method additionally includes receiving, from at least one client by the central model server, information indicative of a respective updated local ML model. The method also includes determining, based on the information indicative of the respective updated local ML models, an updated global ML model.